August Essential Dynamics of Zn Superoxide Dismutase ... · PDF fileIntersubunit Communication ... (Getzoff et al., 1983; Djinovich-Carugo et al., 1994). ... 1992; Romo et al., 1995)

Biophysical Journal Volume 73 August 1997 1007-1018

The Essential Dynamics of Cu, Zn Superoxide Dismutase: Suggestion ofIntersubunit Communication

G. Chillemi,*# M. Falconi,§ A. Amadei,n G. Zimatore,*I1 A. Desideri,5 and A. Di Nola**Department of Chemistry, University of Rome "La Sapienza," 00185 Rome, Italy; #Consorzio per le Applicazioni di Supercalcolo perUniversita e Ricerca, c/o Universita "La Sapienza," 00185 Rome, Italy; §1stituto Nazionale Fiscia Della Materia and Department of Biology,University of Rome "Tor Vergata," 00173 Rome, Italy; 'nReseach Institute and Laboratory of Biophysical Chemistry, University ofGroningen, 9747 AG Groningen, The Netherlands; and Illstituto di Strutturistica Chimica, Consiglio Nazionale Delle Ricerche Montelibretti,Rome, Italy

ABSTRACT A 300-ps molecular dynamics simulation of the whole Cu, Zn superoxide dismutase dimer has been carried outin water, and the trajectory has been analyzed by the essential dynamics method. The results indicate that the motion isdefined by few preferred directions identified by the first four to six eigenvectors and that the motion of the two monomersat each instant is not symmetrical. The vectors symmetrical to the eigenvectors are significantly sampled, suggesting that, onaverage, the motions of the two subunits will exchange. Large intra- and intersubunit motions involving different subdomainsof the protein are observed. A mechanical coupling between the two subunits is also suggested, because displacements ofthe loops surrounding the active site in one monomer are correlated with the motion of parts of the second toward theintersubunit interface.

INTRODUCTION

Cu, Zn superoxide dismutases (SODs; EC 1.15.1.1) areubiquitous metalloenzymes that catalyze the dismutation ofsuperoxide into oxygen and hydrogen peroxide by alternat-ing reduction and oxidation of a copper ion that constitutesthe active redox center (Bannister et al., 1987). The struc-ture-function relationship of Cu, Zn SODs is now quite wellunderstood from solution studies of native and site-directedmutated enzymes (Getzoff et al., 1992; Fisher et al., 1994;Polticelli et al., 1995, 1996) and from the available three-dimensional structures (Tainer et al., 1982; Djinovich et al.,1992; Djinovich-Carugo et al., 1996). The diffusion of thenegatively charged substrate may be modulated by the dis-tribution of the electrostatic potential around the protein(Klapper et al., 1986; Sines et al., 1990), which has beensuggested to be constant in the evolution of this enzyme(Desideri et al., 1992; Sergi et al., 1994).The folding of the amino acid polymer is globular and is

constituted by two identical subunits of 16 kDa, which havebeen called Orange and Yellow, respectively (Tainer et al.,1982). Each subunit is composed of a flattened cylinder(called a (-barrel) and three major external loops, whichform almost half of the subunit (Tainer et al., 1982) (Fig. 1).The (3-barrel structure is constituted by two (3-sheets; one iscomposed of four regular (1-4) and the other by fourtwisted (5-8) ,B-strands joined by three large loops (6, 5-7,8, and 4, 7) and by four turns. In combination with loops 6,5and 7,8, the latter, less regular (3-sheet, accommodates the

Received for publication 17 January 1997 and in final form 8 May 1997.Address reprint requests to Dr. Alessandro Desideri, Department of Biol-ogy, University of Rome "Tor Vergata," Via della Ricerca Scientifica,00133 Rome, Italy. Tel.: + 39-6-72594376; Fax: +39-6-72594311; E-mail:desideri @utovrm.it.( 1997 by the Biophysical Society0006-3495/97/08/1007/12 $2.00

active site. Loop 4,7 joins (3-strands belonging to differentsheets, and the other two large loops (loops 6, 5 and 7, 8)form a deep crevice connecting the solvent and the externalsurface of the barrel. The copper and zinc atoms are locatedat the bottom of this channel and are coupled together by abridging imidazolate side chain (His61). The Zn2+ ion iscoordinated by two more histidyl residues and one aspartylresidue (His69, His78, Asp8l). The catalytically activeCu2+ ion is accessible to the solvent and is coordinated bythree additional histidyl residues (His44, His46, His118)and by one water molecule (Tainer et al., 1982), which is infast exchange with the bulk water (Gaber et al., 1972) andwhich may be substituted by the incoming anion inhibitorsor substrate (Getzoff et al., 1983; Djinovich-Carugo et al.,1994).Some dynamical properties of this protein have been

investigated by experimental and theoretical approaches.The protein average dynamics, worked out through a quasi-elastic neutron scattering study (Andreani et al., 1995),indicated a mean square displacement of the nonexchange-able protons comparable to that observed in myoglobin(Doster et al., 1989). Analysis of the electronic absorptionspectra of the metals as a function of temperature indicateda dynamical coupling with the solvent for the copper but notfor the zinc site (Cupane et al., 1994, 1995). Moleculardynamics (MD) simulations of the bovine protein (Shen etal., 1989; Shen and McCammon, 1991; Y. Wong et al.,1993) and homology models of native and mutant humanCu, Zn SOD (Banci et al., 1992, 1994) pointed out theimportance of the mobility of the active site channel forinteraction with the substrate. All of these MD studies havebeen performed on a simulation system composed of onemonomer and part of the second, including the interfaceregion. Recently a 100-ps MD simulation (Falconi et al.,1996) on the whole dimer suggested the occurrence of two

1 007

Volume 73 August 1997

FIGURE 1 Schematic view of the bovine Cu, Zn superoxide dismutasedimer. The arrows indicate the (3-strands ((3-strands 1-4 form the regular,B-sheet, whereas (3-strands 5-8 form the irregular (3-sheet). The threemajor loops and the four turns are represented as thin wires. The disulfidebridge (S-S) is indicated by the ball-and-stick representation. The spheresrepresent the metal ions. The picture was produced with the MolScript v 1.4program (Kraulis, 1991).

different conformational substates for the two monomers.To better understand such a phenomenon, we decided tocarry out an essential dynamics analysis on this enzyme.

Essential dynamics has been shown to be a very usefultool for finding evidence of the protein regions where themotion is mainly localized, and since its introduction (Ama-dei et al., 1993) it has been successfully applied to describ-ing the dynamical characteristics of several monomeric pro-teins (Amadei et al., 1993; van Aalten et al., 1995; Scheeket al., 1995).

In this study we report for the first time the essentialdynamics analysis of a 300-ps MD simulation of a wholedimeric molecule in water to investigate and highlight theprincipal motions due to the protein dimericity. The resultsindicate an instantaneous asymmetrical dynamical behaviorof the two monomers, the confinement of the largest dis-placements in the region constituted by the active site loops,and the presence of correlated motions that suggest theoccurrence of a mechanical coupling between the twosubunits.

METHODS

Molecular dynamics simulation has been performed withthe parallel version of GROMOS (van Gunsteren and Be-rendsen, 1987) realized under the EUROPORT2 program(Parallelization for Computational Chemistry PACC con-sortium) of the European Commission. Calculations havebeen carried out on a Silicon Graphics Power Challenge (16nodes shared memory) platform.The coordinates of the Orange/Yellow Cu, Zn SOD

dimer at 2-A resolution (Tainer et al., 1982) were obtainedfrom the Brookhaven Protein Data Bank (Bernstein et al.,1977). The dimeric molecule was immersed in a rectangularbox of dimensions 61 X 65 X 82 A3, filled with 9436 watermolecules. For the active site metal cluster, ab initio calcu-lated atomic partial charges were used (Shen et al., 1990).

Nonpolar hydrogen atoms were not included in the calcu-lation, and the united atom approach was, applied. Theprotein, including polar hydrogen atoms, contained 2678atoms. A total of 158,000 steps of 2 fs were performed,reaching a total simulation time of 316 ps. The CPU time forthe simulation of each step was 2.2 s.The protocol for the equilibration was the following:1. Ten picoseconds of solvent equilibration with the pro-

tein atoms fixed2. Ten-picosecond simulation of the whole system, start-

ing at a temperature of 10 K and ending at a temperature of298 K

3. Simulation (316 ps) at T = 298 K.The first 50 ps of the simulation was used for equilibra-

tion, and the remaining 266 ps was used for analysis. Thetemperature was kept constant at 298 K by coupling thesystem to an external bath (Berendsen et al., 1984). Bondlengths were constrained by the SHAKE procedure (Ryck-aert et al., 1977), and nonbonded interactions were evalu-ated using a cutoff of 8.0 A. Protein and solvent configu-rations were saved every 50 steps, so that 3160 frames werestored. The trajectory was checked to assess the quality ofthe simulation, and the essential degrees of freedom wereextracted according to the method developed by Amadei(Amadei et al., 1993) by using the program WHATIF(Vriend, 1990). The method, equivalent to a principal com-ponent analysis on the coordinate fluctuations (Garcia,1992; Romo et al., 1995) and related to quasi-harmonicanalysis (Ichiye and Karplus, 1991; Hayward et al., 1993),allows separation of the configurational space into sub-spaces: an essential subspace, in which most of the posi-tional fluctuations can be described by a few coordinates,and a remaining subspace including all of the coordinateswith an approximate constrained harmonic character. Theessential dynamics method is based on the construction ofthe correlation matrix of the coordinate fluctuations:

C = cov(x) = ((X- (x))(x- (X))T) (1)

After removal of the translational and rotational degreesof freedom, the matrix C was built up and then diagonalizedto obtain the eigenvectors and eigenvalues, which giveinformation about correlated motions throughout the pro-tein. By ordering the eigenvectors, according to decreasingvalue of the eigenvalues, the most relevant eigenvectors canbe detected. In the present case, because of the large size ofthe macromolecule investigated, the correlation matrix wasbuilt up only for the protein C,, atoms. It has already beenshown that this choice gives a correct picture of the essentialsubspace (Amadei et al., 1993).

RESULTS

Structural check

Geometrical properties of Cu, Zn superoxide dismutasewere monitored during the whole trajectory as a measure of

1 008 Biophysical Journal

Essential Dynamics of Cu, Zn Superoxide Dismutase

structural stability. This check was done iteratively with theDSSP program (Kabsch and Sander, 1983), using the mo-

lecular dynamics frames as input and by comparing themwith the minimized x-ray structure.To control the structural stability of the protein, the

following geometrical properties have been monitored: thetotal accessible surface area of the protein dimer, the num-

ber of residues in unfavorable regions of the Ramachandranplot, the number of residues in random coil conformation,the number of backbone hydrogen bonds, the gyration ra-

dius for the C. structures, the percentages of secondarystructure, and the r.m.s.d. from the minimized x-ray struc-ture in each SOD monomer. The time dependences of thesevariables are reported in Fig. 2 and 3. For the sake of clarity,both the averages and the root mean square fluctuations(r.m.s.f.) are reported in Table 1, in comparison with thecorresponding values observed in the minimized crystalstructure. Fig. 2 shows that the largest variation in theparameters, with respect to the crystallographic structure,occurred within the first 50 ps. The structural parameters b,c, d, and e in Fig. 2 do not change appreciably during theremaining simulation, whereas the total accessible surface

60.0

, 50.00

cI 40.0

a)c 30.00

a)U').o 20.0

60.0

, 50.0cJ

40.0

a)cn10c 30.00

O

a)

Cf)20.0

10.0

4.0

.z 3.0

-6

2.0

1.0

0.017000.0

160000

15000.0

14000.0

50

40

30

140130

120110

10090

220

200

180

160

140

19.50

19.00

18.50

18.000 50 100 150

Time (ps)

FIGURE 2 Time dependence of SOD geomeaccessible surface area of the protein dimer (Ain unfavorable regions of the Ramachandran plin random coil conformation. (d) Number of ba(e) Gyration radius for the C, structures (A).

150Time (ps)

FIGURE 3 Time dependence of the percentage of ,B ( ) and random-coil (.) structure in the orange (a) and yellow (b) SOD monomers. (c)Time course of the all-atom r.m.s. deviation from the minimized x-ray

structure of the orange ( ) and the yellow (.) subunits.

area of the dimer reaches a constant value in the second halfof the simulation. The increase in the total accessible sur-

face area and the constancy of the other parameters indicate-±-|---+ 1- 1 1 1 1 that the protein maintains its structure and that the fluctua-

tions are mainly localized on the loops.The percentage of secondary structure, calculated for

each subunit shown in Fig. 3, is indicative of asymmetricalmotion in the two monomers. In the Orange subunit thepercentage does not change during the trajectory. In theYellow subunit the percentage of the ,B structure is reducedslightly, and there is a corresponding increase in the randomcoil. It has been shown that the ,B-sheet structure can easilybe deformed during dynamics simulations (Sneddon andBrooks, 1993). Sheets can undergo backbone dihedral os-

cillations without losing hydrogen bond energy due to thecurvature of the - potential energy surface that issmaller in the /3-region than in the a-region. In the case ofSOD, this alteration does not perturb the correct globularfolding, as confirmed by the total number of main-chain

200 250 300 hydrogen bonds that is maintained in the whole dimer (seeTable 1). The asymmetry of the two monomers is alsoconfirmed by the time course of the all-atom r.m.s.d. from

2trical properties (a) Total the minimized x-ray structure of the two separate subunitslot. (c) Number of residues (Fig. 3 c).ackbone's hydrogen bonds. The exchange of water molecules in the active site has

been analyzed for each subunit as a function of time. In the

a

C

d I. 1,eChillemi et al. 1 009

Volume 73 August 1997

TABLE I Geometrical properties and fluctuations

Geometrical property Reference MD average r.m.s.f.

Total solvent accessible surface area of the dimer (A2) 13755 15525 581Residues outside the allowed Ramachandran regions 49.0 34.7 3.6Residues in random coil conformation 95 115 9Main-chain to main-chain hydrogen bonds 188 176 10Gyration radius of Ca^ atoms (A) 19.6 18.9 0.1% of 3 structure in Orange subunit 35.10 28.78 3.56% of random coil structure in Orange subunit 28.48 34.62 2.91% of (3 structure in Yellow subunit 28.00 25.22 3.79% of random coil structure in Yellow subunit 34.67 42.01 4.46

Yellow subunit, one water molecule remains close to thecopper (average Cu-O distance 3.0 A), whereas in theOrange subunit four water molecules (average Cu-O dis-tance 2.9 A) exchange with the bulk and alternately ap-proach the copper atom.

Essential dynamics analysis

Fig. 4 shows a plot of the eigenvector index against eigen-values, derived from the covariance matrix obtained fromthe solvent simulation trajectory.

There are only a few eigenvectors with large eigenvalues,as already reported for lysozyme (Amadei et al., 1993),thermolysin (van Aalten and Amadei, 1995), and HPr(Scheek et al., 1995), showing that the protein motionoccurs mainly along very few directions in the essentialsubspace. The fractional contribution to the overall motionappears as an inset in the figure.

300.0

200.0

FIGURE 4 Plot of eigenvaluesagainst corresponding eigenvector in-dices derived from the covariancematrix constructed from the simula-tion. The eigenvectors are sorted bydecreasing eigenvalue. (Inset) Frac-tional contribution to the overallmotion.

01)C)wr

100.0

0.00

The components of the first four eigenvectors are re-ported in Fig. 5 for each monomer. From the figure it isevident that the amplitude of the essential motions of thetwo monomers is different, indicating the occurrence of aninstantaneous asymmetry. In particular, in the first twoeigenvectors, a larger motion is observed in the Yellowmonomer than in the Orange one. The opposite occurs foreigenvectors 3 and 4, in which the Yellow monomer is lessmobile than the Orange one. It is also interesting to note thatin both monomers the largest displacements occur in theresidues belonging to the electrostatic loop 7,8, which isinvolved in the substrate steering toward the catalytic cop-per (Klapper et al., 1986; Sines et al., 1990). It can be notedthat each eigenvector has components of nonnegligible in-tensity in both subunits. This is particularly evident forloops 7,8 and 6,5. This is evidence of a correlation betweenthe subunits and suggests the presence of a mechanicalcoupling between them. In Fig. 6 the projections of the Ca

40 60Eigenvector index

1010 Biophysical Journal

Essential Dynamics of Cu, Zn Superoxide Dismutase

Eigenvector 1

Component number

Eigenvector 20.20

0.15

0.10

0.05

0.00 L

Eigenvector 3

I 1-.--1000 100Component number Component number

0.20

0.15

0.10

0.05

0.00

Eigenvector 4

Component number

Eigenvector 10.20 '

~~I0.15 1-

0.10 I I

I I

I I

0.05 -I

II

0.00152 252

Component number

0.20

0.15

0.10

0.05

Eigenvector 2

0.00 as.' ;I I M152 252

Component number

0.20

0.15

0.10

0.05

Eigenvector 3

0.00 L II I

152 252Component number

Eigenvector 4

Component number

FIGURE 5 Absolute value of the components of the first four eigenvectors for each SOD subunit (absolute values). The four upper graphs represent the

orange subunit, and the four lower graphs represent the yellow subunit. Boundaries of the loop regions: , loop 6, 5; - - -, loop 4, 7;., loop 7, 8.

trajectory on selected eigenvectors are shown. The motion"period" of the first eigenvector is larger than the time range

of the simulation. Eigenvectors 2, 3, and 4 show an almost"cyclic" motion. Starting from eigenvector 10, no relevantmotion can be detected, and only Gaussian oscillationsaround an average value are observed.

Instantaneous asymmetrical motion of thetwo monomers

Fig. 5 indicates that the dynamic behavior of the two mono-

mers is instantaneously asymmetrical, whereas in a proteinmade of two identical subunits, associated in a symmetricalway, we should expect that the atomic motions are invariantwith respect to the symmetry of the dimer. This does notimply that at each time the molecule should be symmetrical,but that every type of motion and its "symmetric" (obtainedby applying the symmetry operation, which exchanges thesubunits) should have identical average properties. Thesymmetrical vector corresponding to each eigenvector maybe obtained by applying the symmetry operation on theeigenvector and is not expected to be another eigenvector,because symmetrical vectors are not generally orthogonal inthe configurational space. In this case, application of thesymmetry operation to a given eigenvector generates a

vector that can be expressed as a linear combination of theeigenvectors.To check if the essential subspace has been significantly

sampled, we have analyzed the motion along the vectorssymmetrical to the eigenvectors.

In Fig. 7 we show the average square fluctuations alongvectors symmetrical to the first 500 eigenvectors. From thefigure it is evident that most of the fluctuation is confined tothe first 100 vectors and that this curve is highly correlatedwith the curve of Fig. 4, which reports the eigenvectorsagainst their relative eigenvalues. In particular, the first10-15 symmetrical vectors have the highest fluctuationvalues, which implies that a simulation of -300 ps producesa significant sampling of these directions. In particular, theirvalues are comparable in size to the eigenvalues corre-

sponding to the eigenvectors of indices in the range 5-10. InFig. 8 we show the cumulative square projections of vectors1-4, 50, and 250 symmetrical to the corresponding eigen-vectors onto the first 500 eigenvectors. Vectors 1-4 are

largely defined within the first 50 eigenvectors, whereasvectors 50 and 250 can reach a comparable cumulativesquare projection only within 150 and 250 eigenvectors,respectively.

These results imply that our simulation provides a reli-able sampling of the phase space and gives a good indica-

0.20

) 0.15.(aC0)C 0.100

E0o 0.05

0.00

a-0

0

-

._NCc

0CLE00

1011Chillemi et al.

Volume 73 August 1997

Eigenvector 130.0

20.0

p 10.0

0.0

-10.0

-20.0100 200

Eigenvector 2 Eigenvector 330.0

Eigenvector 4

20.0 V

10.0

0.0

-10.0

I ,I '-20.00 100 200 0 100 200

Eigenvector 5

Time (ps)

Eigenvector 10

0 100 200Time (ps)

30.0

20.0 F

10.0

0.0

Eigenvector 20

-10.0

-20.0 L0 100 200

Time (ps)

30.0Eigenvector 50

20.0 F

10.0

0.0

-10.0

-20.0 L0 100 200

Time (ps)

FIGURE 6 Projections of the simulation frames on selected eigenvectors.

tion that although the motion of the dimer is instantaneouslyasymmetrical it can be expected that on the average themotions of the two subunits will exchange.

Structural motions

To better localize the main regions involved in the proteinmotions, the components of the first four eigenvectors have

20.0

15.0

.S0

IR10.0

CUD

5.0

0.0

been represented as individual 3D vectors originating fromthe Ca atoms of the time-averaged structure of the protein.This "C,-arrows" representation is useful in understandingthe localization of the protein motions for each eigenvector.It must be pointed out that the arrows indicate the directionof the motion, whereas the intensity and versus can change,only coherently and simultaneously, for all of the arrows.

1.0

0.8

0

.a,0.6

E

0)

0.2

0.00 100 200 300

Eigenvector index400 500

Vector index

FIGURE 7 Amount of motion (average square fluctuation) along vectors

symmetrical with respect to the eigenvectors.

FIGURE 8 Cumulative square projections of the vectors symmetrical tothe eigenvectors onto the first 500 eigenvectors. , vector 1;., vector2; -, vector 3; -, vector 4; vector 50; vector 250).

30.0

DE

.co0Co

20.0

10.0

0.0

-10.0

-20.0 L0

a)E

co8CUCO

c

1012 Biophysical Journal

-

Essential Dynamics of Cu, Zn Superoxide Dismutase

The motions observed are mainly concentrated in two largeregions of the protein subunit. The first, of higher intensity,involves the subdomains including the major loops of SOD(6, 5 and 7, 8); the second, of lower intensity, involves thesubdomains forming the antiparallel ,3-barrel. The structuralfluctuations corresponding to each eigenvector are gener-ated by a different combination of the motions involving theloops and the 13-barrel, and the resulting global motion ofeach eigenvector has an asymmetrical character.The motions evidenced by the first eigenvector are re-

ported in Fig. 9. These motions are mainly concentrated onthe Zn-ligand region of loop 6,5 and the electrostatic loop7,8, whereas the ,8-barrels are less mobile than the loops.The direction of the motion of loop 7,8 is different in thetwo subunits, pointing alternately toward the solvent andtoward the protein in the Orange and Yellow subunits,respectively. The resulting motion in the two subunits isasymmetrical, in both direction and intensity, being ofhigher intensity in the Yellow compared to the Orangesubunit (see also Fig. 5). Loop 4,7, which is the only loopconnecting the two 13-sheets in each monomer, simulta-neously points toward or away from the dimer interface. Inthe protein orientation of Fig. 9, loop 4,7 is not clearlyshown.

Analysis of this eigenvector indicates that the 13-barrelacts as a framework into which loops 6,5 and 7,8 areinserted, and these are free to fluctuate, modifying the shapeof the active site. In SOD at physiological conditions, therate-limiting step of the catalytic reaction is the collisionbetween the substrate and the catalytic site, which is knownto be electrostatically controlled (Desideri et al., 1992; Sergiet al., 1994). The motion observed in this eigenvector pro-duces a reorganization of the loops constituting the electro-static channel to enhance the interaction with the substrate.

The motions corresponding to the second eigenvector arerepresented in Fig. 10 with the same protein orientation ofthe previous picture. The motion, as for the previous eig-envector, is of higher intensity in the Yellow (see also Fig.5) than in the Orange subunit. In the Yellow subunit, theregular 1-sheet of the 13-barrel and the major loops 6,5 and7,8 move coherently in opposite directions. At the sametime, there is a small-intensity motion of the Orange subunittoward and away from the Yellow one (Fig. 10). Loop 4,7is characterized by a motion similar to that described for thefirst eigenvector.The Ca components of the third eigenvector are shown in

Fig. 11. The orientation of the protein is different from thatshown in Figs. 9 and 10. The motion may be divided intotwo circular movements of opposite direction, each involv-ing one subunit. In this representation the 13-barrel frame-work pushes toward loops 6,5 and 7,8 of the same subunit,twisting loop 4,7 downward, moving it toward the interface.The motions of higher intensity are localized in loops 6,5and 7,8 of the Orange subunit (see Fig. 5). It must bepointed out that the direction of motion may be reversed.The C. components of the fourth eigenvector are repre-

sented on the dimer in Fig. 12 in the same orientation as inFigs. 9 and 10. The motions are mainly concentrated onloops 6,5 and 7,8 of the Orange subunit (see also Fig. 5).The direction of motion follows the external arrows repre-sented in the figure. The Yellow 6,5 and 7,8 loops move,with low intensity, toward the 13-barrel of the same subunitand then toward the dimer interface. The motion transmittedto the Orange 13-barrel is finally released to external loops6,5 and 7,8 of the same subunit, which move outward in thesolvent. In this asymmetrical motion, the dynamic couplingbetween the two subunits is correlated with the large mo-tions on the Orange loops, which result in a fluctuation of

FIGURE 9 The motions of the first eigenvector are represented by arrows associated with each C,a atom on a ball-and-stick stereo representation of theSOD dimer. The copper ion is represented by a dark sphere, and the zinc ion is shown as a light gray sphere. The orange and the yellow monomers arerepresented in the upper right and in the lower left parts of the picture, respectively.

Chillemi et al. 1013

Volume 73 August 1997

FIGURE 10 Same as Fig. 9, but for the second eigenvector.

residues located at the active site border, which must beflexible to facilitate the enzyme-substrate interaction.

Structural perturbation in the linearresponse limit

In the linear response limit, the covariance matrix can yeldinsights into the structural response of the molecule whensmall perturbations are introduced. This can be obtained bythe Green's function/principal component analysis (cf.Wong et al., 1993). If a set of small forces is applied to theatoms of the molecule, the change in the average atomicpositions can be obtained from

A(x) = (3CFP (2)

where 1B = llkT and FP are the perturbation forces, and Crepresents the correlation matrix. From this equation itfollows that when a force Pi, such that 1BFPiX = 1, is applied

to a single atom, i, in the x direction, its effect on atom j is

ARjX = + A(y)2X + A(Zj)2= (XiAXj)2 + (AXiAyj)2 + (AXiAzj)2

(3)

A similar displacement of atom j occurs when the sameforce is applied on atom i along the y and z directions. Toobtain a single value from AR, ARJ , and ARjiZ thefollowing averaging procedure has been carried out:

(ARiX + AM + R2i 1/2

(4)

In Fig. 13 we report the matrix aji = ARji, normalized tothe maximum aji value. The j and i index represents the Ca,of the jth and ith residues, respectively. The grey and blackpixels correspond to 0.2 ' aji ' 0.6 and 0.6 ' aji ' 1.0,respectively, whereas no pixel is recorded for aji ' 0.2.

In the matrix, the lower left (IH) and upper right (III)panels show the intrasubunit correlations in the Orange and

FIGURE 11 Same as Fig. 9, but for the third eigenvector. The protein is represented in a different orientation in which the orange and yellow monomers

are displayed in the right and left parts of the picture, respectively.

1014 Biophysical Journal

Essential Dynamics of Cu, Zn Superoxide Dismutase

I /

//

FIGURE 12 Same as Fig. 9, but for the fourth eigenvector. The external arrows indicate the direction of the motion observed in this eigenvector.

Yellow subunits, respectively. The two panels confirm thatthe motion in the two subunits is asymmetrical, inasmuch asthe Yellow monomer shows an high degree of internalcorrelation if compared with the Orange subunit. It is inter-esting, however, that in both subunits the loop 7,8, whichconstitutes part of the electrostatic channel that modulatesthe collision of the substrate with the active site (Fisher etal., 1994; Polticelli et al., 1995), is entirely correlated witha large number of residue groups scattered along the proteinsequence.

Interesting correlations are also observed between thetwo subunits (I and IV) where, in the same way, the elec-trostatic loop 7,8 and parts of the loops 4,7 and 6,5, includ-ing the region involved in the Zn binding, show correlationswith several residues of the other subunit. These resultsconfirm the data presented in the previous section andprovide further evidence of the crucial role of loops 6,5 and7,8, which are known to be involved in the substrate steer-ing toward the catalytic copper atom (Sines et al., 1990;Polticelli et al., 1995; Fisher et al., 1994).

DISCUSSION

In the present paper we have reported the results of a 300-pssimulation of the dimer of SOD in solution. This is thelongest simulation of the whole dimer in solution reportedso far; however, it is still relatively short, and we arecertainly far from full equilibration and sampling of thephase space accessible to the molecule. Nevertheless previ-ously reported results (Amadei et al., 1993, 1996; vanAalten et al., 1995, 1996; Scheek et al., 1995; de Groot etal., 1996a,b,c) on proteins in solution have shown that theessential subspace of a protein can be reasonably deter-mined by a simulation of a few hundred picoseconds. This

is a crucial point in assessing the reliability of the presentresults. Analysis of the vectors symmetrical to the firsteigenvectors supports the idea that the essential subspacehas been significantly sampled. As the dimer is symmetri-cal, it is to be expected that on the average, an eigenvectorand its symmetrical vector have the same amplitude. Figs. 7and 8 show that essential subspace of symmetrical vectorscorresponds to the essential subspace of eigenvectors; i.e.,the first symmetrical vectors correspond to the first eigen-vectors. We can conclude that, although a much longersimulation is required for an exhaustive sampling and sta-tistics, the present simulation and analysis have likely de-tected the main features of the SOD dimer in solution.The motions described in the four eigenvectors indicate

that the two subunits behave at each instant in an asymmet-rical way. This asymmetrical behavior is also evidenced by:1) an r.m.s.d. different from that of the x-ray minimizedstructure (Fig. 3); 2) the different time exchange of thewater molecule close to the copper site in the two subunits;3) the projection of the motions on each Ca (Fig. 5); 4) thematrix shown in Fig. 13. It is important to notice that anindependent MD study on the same protein (Falconi et al.,1996), carried out under different conditions, has also re-ported the occurrence of an asymmetrical dynamic for thetwo subunits.

Analysis of the second and the fourth eigenvectors sug-gests that concerted motions of large regions of one subunit(the loops and the ,3-barrel) influence the movement of theother subunit, leading to the hypothesis that the asymmetrybetween the two SOD monomers may be due to this inter-subunit communication. The suggestion of an intersubunitcommunication also comes from the matrix of Fig. 13,which shows a correlation between several residues of onesubunit and loop 7,8, loop 6,5, and loop 4,7 of the other

1015Chillemi et al.

Volume 73 August 1997

6;f '6 .yr.4 55a

r, l.: fX tNP 45 6rse 66|X vX

/66~~~~~~~~~~/"*,,e esw-i

PIkW:g 4,,1. 9,1-0O 'll,R66P

>.f .N#S9,y. . ;S. f

py>

AL.M56.

U, WIW jB* WI i t:dfif4/O > ]Bw w gp @ X MA < >

C.P,I. VI

p"X^--4g .0 614

iif,/,V VI' ly frf

sW WW3&tIt W; W-

200

a s §lON

x xx o

250

FIGURE 13 Displacement correlation matrix aj* as defined in Eq. 4. The grey and black pixels coffespond to 0.2 ' aji ' 0.6 and 0.6 ' aj4 ' 1.0,respectively, whereas no pixel is reported when aji ' 0.2. The upper-left square (I) and the lower-right square (IV) indicate the intersubunit correlations.The lower-left square (IH) and the upper-right square (III) indicate the intrasubunit correlations in the orange and yellow subunits, respectively.

subunit. Loop 7,8 and loop 6,5 are the two major loops ofSOD, which, together with the less regular (3-sheet, accom-modate the active site, and their motion is needed to orga-nize the surrounding of the active site in a conformation thatmay allow the specific interaction of the catalytic copperwith the superoxide. Previous MD studies (Shen et al.,1989; Shen and McCammon, 1991; Falconi et al., 1996)have indicated that the superoxide is directed toward thecatalytic metal through the concomitant motions of severalinvariant residues located in the proximity of the active site.The significant role of protein fluctuations in the enzyme-substrate interactions has been confirmed by a study com-

bining the use of Brownian and molecular dynamics in thepresence of the substrate (Luty et al., 1993) and by MDcalculations carried out on homology models of human Cu,Zn SOD mutants (Banci et al., 1992, 1994).The simulation of the whole dimer in combination with

the essential dynamics analysis allows us to highlight, in

addition to the motion of the active site loops, the presenceof correlated motions in the two subunits that suggest a

mechanical coupling between the two subunits. This resultis in agreement with the experimental evidence of a con-

formational communication between the two remote coppersites of the native dimer, detected as a break at one equiv-alent per mole of protein in either coulometric titration(Lawrence and Sawyer, 1979) or in the process of metalreconstitution of the apoprotein (Rigo et al., 1977, 1978).

Concerning the instantaneous asymmetrical behavior ofthe two subunits, it should be noted that from Figs. 7 and 8we can conclude that our simulation provides a reliablesampling of the motion of the symmetrical vectors, suggest-ing that, over time, an exchange of the two subunits shouldoccur. Our simulation cannot indicate the time needed toproduce such an exchange, and work is in progress tolengthen the trajectory and estimate such a value. Becausethe second-order catalytic rate constant of this enzyme is

.1.

I'll

300

250

200

721* Al

6z-

AN

t w ,RB

A, 44E

a)

C)a/)

+

46

I..Cf>st

OWNWoolV.

R.N0f.f

,WM i.fEf. -,3

150

5.M

Y;% -:

NOfx

:.,

v 6 45W1

Z ,fsegA 4

III

IV

100

11

50

00 50 100 150

Sequence300

T-I T- I-- I- IT I-r-- -- f-T- T--Y------ --

o

IL t A .1

1016 Biophysical Journal

I

Chillemi et al. Essential Dynamics of Cu, Zn Superoxide Dismutase 1017

limited by the diffusion (kcat/lKM 0109 M 1 s- 1), determi-nation of the time needed to exchange the motion of the twosubunits would be of great importance to understandingwhether the substrate meets a symmetrical or an asymmet-rical enzyme, i.e., if both subunits work at the same time orin an alternate way. Up to now contradictory experimentalresults have been reported in the literature (Fielden et al.,1974; Cockle and Bray, 1977; Viglino et al., 1981).

The authors acknowledge the Ente Nazionale Energie Alternative groupresponsible for the benchmarking activity of the code within the Europort2/PACC project for making the simulation data available.

This work was partially supported by the Italian National Research Councilspecial project "Sviluppo di algoritmi innovativi per la simulazione staticae dinamica." We thank CNR Italy for the financial support of G. Zimatore.

REFERENCES

Amadei, A., A. B. M. Linssen, and H. J. C. Berendsen. 1993. Essentialdynamics of proteins. Proteins Struct. Funct. Genet. 17:412-425.

Amadei, A., A. B. M. Linssen, B. L. de Groot, D. M. F. van Aalten, andH. J. C. Berendsen. 1996. An efficient method for sampling the essentialsubspace of proteins. J. Biomol. Struct. Dyn. 13:615-625.

Andreani, C., A. Filabozzi, F. Menzinger, A. Desideri, A. Deriu, and D. DiCola. 1995. Dynamics of hydrogen atoms in superoxide dismutase byquasi-elastic neutron scattering. Biophys. J. 68:2519-2523.

Banci, L., P. Carloni, G. La Penna, and P. L. Orioli. 1992. Moleculardynamics studies on superoxide dismutase and its mutants: the structuraland functional role of Argl43. J. Am. Chem. Soc. 114:6994-7001.

Banci, L., P. Carloni, and P. L. Orioli. 1994. Molecular dynamics studieson mutants of Cu, Zn superoxide dismutase: the functional role ofcharged residues in the electrostatic loop VII. Proteins Struct. Funct.Genet. 18:216-230.

Bannister, J. V., W. M. Bannister, and G. Rotilio. 1987. Aspects of thestructure, function and applications of superoxide dismutase. CRC Crit.Rev. Biochem. 22:111-180.

Berendsen, H. J. C., J. P. M. Postma, W. F. van Gunsteren, A. Di Nola, andJ. R. Haak. 1984. Molecular dynamics with coupling to an external bath.J. Chem. Phys. 81:3684-3690.

Bernstein, F., T. Koetzle, G. Williams, E. Meyer, Jr., M. Brice, J. Rodgers,0. Kennard, T. Shimanouchi, and M. Tasumi. 1977. The protein databank: a computer-based archival file for macromolecular structures.J. Mol. Biol. 112:535-542.

Cockle, S. A., and R. C. Bray. 1977. Do all the copper atoms in bovinesuperoxide dismutase function in catalysis? In Superoxide and Super-oxide Dismutase. A. M. Michelson, J. M. McCord, and I. Fridovich,editors. Academic Press, London. 215.

Cupane, A., M. Leone, V. Militello, M. E. Stroppolo, F. Polticelli, and A.Desideri. 1994. Low-temperature optical spectroscopy of native andazide reacted bovine Cu, Zn superoxide dismutase. A structural dynam-ics study. Biochemistry. 33:15103-15109.

Cupane, A., M. Leone, V. Militello, M. E. Stroppolo, F. Polticelli, and A.Desideri. 1995. Low-temperature optical spectroscopy of cobalt in Cu,Co superoxide dismutase. A structural dynamics study of the solvent-unaccessible metal site. Biochemistry. 34:16313-16319.

de Groot, B. L., A. Amadei, and H. J. C. Berendsen. 1996a. Towards anexhaustive sampling of the configurational spaces of the two forms ofthe peptide hormone Guanylin. J. Biomol. Struct. Dyn. 13:741-751.

de Groot, B. L., A. Amadei, R. M. Scheek, N. A. J. van Nuland, and H. J.C. Berendsen. 1996b. An extended sampling of the configurational spaceof HPr of E. coli. Proteins Struct. Funct. Genet. 26:314-322.

de Groot, B. L., D. M. F. van Aalten, A. Amadei, and H. J. C. Berendsen.1996c. The consistency of large concerted motions in proteins in mo-le,cular dynamics .simulatio ns. Biophys. Jt. 71l:170)7-1713.

Desideri, A., M. Falconi, F. Polticelli, M. Bolognesi, K. Djinovic, and G.Rotilio. 1992. Evolutionary conservativeness of electric field in the Cu,Zn superoxide dismutase active site. J. Mol. Bio. 223:337-342.

Djinovic, K., G. Gatti, A. Coda, L. Antolini, G. Pelosi, A. Desideri, M.Falconi, F. Marmocchi, G. Rotilio, and M. Bolognesi. 1992. Crystalstructure of yeast Cu,Zn enzyme superoxide dismutase. Crystallographicrefinement at 2.5 A resolution. J. Mol. Biol. 225:791-890.

Djinovic-Carugo, K., A. Battistoni, M. T. Carri, F. Polticelli, A. Desideri,G. Rotilio, A. Coda, and M. Bolognesi. 1994. Crystal structure of thecyanide-inhibited Xenopus laevis Cu, Zn superoxide dismutase at 98K.FEBS Lett. 349:93-98.

Djinovic-Carugo, K., A. Coda, A. Battistoni, M. T. Carri, F. Polticelli, A.Desideri, G. Rotilio, K. S. Wilson, and M. Bolognesi. 1996. Three-dimensional structure of Xenopus laevis Cu, Zn superoxide dismutase Bdetermined by x-ray crystallography at 1.5 A resolution. Acta Crystal-logr. D52:176-188.

Doster, W., S. Cusack, and W. Petri. 1989. Dynamical transition ofmyoglobin revealed by inelastic neutron scattering. Nature. 337:754-756.

Falconi, M., R. Gallimbeni, and E. Paci. 1996. Dimer asymmetry insuperoxide dismutase studied by molecular dynamics simulation.J. Comput. Aided Mol. Des. 10:490-498.

Fielden, E. M., P. B. Roberts, R. C. Bray, D. J. Lowe, G. N. Mautner, G.Rotilio, and L. Calabrese. 1974. The mechanism of action of superoxidedismutase from pulse radiolysis and electron paramagnetic resonance.Evidence that only half the active sites function in catalysis. Biochem. J.139:49-60.

Fisher, C. L., D. E. Cabelli, J. A. Tainer, R. A. Hallewell, and E. D.Getzoff. 1994. The role of arginine 143 in the electrostatic and mecha-nism of Cu, Zn superoxide dismutase: computational and experimentalevaluation by mutational analysis. Proteins Struct. Funct. Genet. 19:24-34.

Gaber, B. P., R. D. Brown, S. H. Koening, and J. A. Fee. 1972. Nuclearmagnetic relaxation dispersion in protein solutions. Bovine erythrocytesuperoxide dismutase. Biochim. Biophys. Acta. 271:1-5.

Garcia, A. E. 1992. Large-amplitude nonlinear motions in proteins. Phys.Rev. Lett. 68:2696-2699.

Getzoff, E. D., D. E. Cabelli, C. L. Fisher, H. E. Parge, M. S. Viezzoli, L.Banci, and R. A. Hallewell. 1992. Faster superoxide dismutase mutantsdesigned by enhancing electrostatic guidance. Nature. 358:347-351.

Getzoff, E. D., J. A. Tainer, P. K. Weiner, P. A. Kollman, J. S. Richardson,and D. C. Richardson. 1983. Electrostatic recognition between superox-ide and copper zinc superoxide dismutase. Nature. 306:287-290.

Hayward, S., A. Kitao, F. Hirata, and N. Go. 1993. Effect of solvent oncollective motions in globular proteins. J. Mol. Biol. 234:1207-1217.

Ichiye, T., and M. Karplus. 1991. Collective motions in proteins; a covari-ance analysis of atomic fluctuations in molecular dynamics and normalmode simulations. Proteins. 11:205-217.

Kabsch, W., and C. Sander. 1983. Dictionary of protein secondarystructure: pattern recognition of hydrogen-bonded and geometrical fea-tures. Biopolymers. 22:2577-2637.

Klapper, I., R. Hagstrom, R. Fine, K. Sharp, and B. Honig. 1986. Focusingof electric fields in the active site of Cu-Zn superoxide dismutase: effectsof ionic strength and amino-acid modification. Proteins Struct. Funct.Genet. 1:47-59.

Kraulis, P. J. 1991. MOLSCRIPT: a program to produce both detailed andschematic plots of protein structures. J. Appl. Crystallogr. 24:946-950.

Lawrence, G. D., and D. T. Sawyer. 1979. Potentiometric titrations andoxidation-reduction potentials of manganese and copper-zinc superoxidedismutases. Biochemistry. 18:3045-3050.

Luty, B. A., S. El Amrani, and J. A. McCammon. 1993. Simulation of thebimolecular reaction between superoxide and superoxide dismutase:synthesis of the encounter and reaction steps. J. Am. Chem. Soc. 115:11874-11877.

Polticelli, F., A. Battistoni, P. O'Neill, G. Rotilio, and A. Desideri. 1996.Identification of the residues responsible for the alkaline inhibition ofCu,Zn superoxide dismutase: a site-directed mutagenesis approach. Pro-tein Sci. 5:248-253.

Polticelli, F., G. Bottaro, A. Battistoni, M. T. Carri, K. Djinovic-Carugo,M. Bolognesi, P. O'Neill, G. Rotilio, and A. Desideri. 1995. Modulationof the catalytic rate of Cu, Zn superoxide dismutase in single and double

1018 Biophysical Journal Volume 73 August 1997

mutants of conserved positively and negatively charged residues. Bio-chemistry. 34:6043-6049.

Rigo, A., P. Viglino, M. Bonori, D. Cocco, L. Calabrese, and G. Rotilio.1978. The binding of copper ions to copper-free bovine superoxidedismutase. Kinetic aspects. Biochem. J. 169:277-280.

Rigo, A., P. Viglino, L. Calabrese, D. Cocco, and G. Rotilio. 1977. Thebinding of copper ions to copper-free bovine superoxide dismutase.Copper distribution in protein samples recombined with less than stoi-chiometric copper ion/protein ratios. Biochem. J. 161:27-30.

Romo, T. D., J. B. Clarage, D. C. Sorensen, and G. N. Phillips, Jr. 1995.Singular value decomposition analysis of time-averaged crystallographicrefinement. Proteins. 22:311-321.

Ryckaert, J. P., G. Ciccotti, and H. J. C. Berendsen. 1977. Numericalintegration of the cartesian equations of motions of a system withconstraints: molecular dynamics of N-alkanes. J. Comp. Phys. 23:327-341.

Scheek, R. M., N. A. J. Van Nuland, B. L. De Groot, A. B. M. Lissen, A.Amadei. 1995. Structure from NMR and molecular dynamics: distancerestraining inhibits motion in the essential subspace. J. Biomol. NMR.6:106-111.

Sergi, A., M. Ferrario, F. Polticelli, P. O'Neill, and A. Desideri. 1994.Simulation of superoxide-superoxide dismutase association rate for sixnatural variants. Comparison with the experimental catalytic rate.J. Phys. Chem. 98:10554-10557.

Shen, J., and J. A. McCammon. 1991. Molecular dynamics simulation ofsuperoxide interacting with superoxide dismutase. Chem. Phys. 158:191-198.

Shen, J., S. Subramaniam, C. F. Wong, and J. A. McCammon. 1989.Superoxide dismutase: fluctuations in the structure and solvation of theactive site channel studied by molecular dynamics simulation. Biopoly-mers. 28:2085-2096.

Shen, J., C. F. Wong, S. Subramaniam, T. A. Albright, and J. A. McCam-mon. 1990. Partial electrostatic charges for the active center of Cu, Znsuperoxide dismutase. J. Comp. Chem. 11:346-350.

Sines, J. J., S. A. Allison, and J. A. McCammon. 1990. Point chargedistributions and electrostatic steering in enzyme/substrate encounter:Brownian dynamics of modified copper/zinc superoxide dismutases.Biochemistry. 29:9403-9412.

Sneddon, S. F., and C. L. Brooks, III. 1993. Protein motions: structural andfunctional aspects. In Molecular Structures in Biology. R. Diamod, T. F.Koetzle, K. Prout, and J. S. Richardson, editors. Oxford UniversityPress, Oxford. 115-163.

Tainer, J. A., E. D. Getzoff, K. M. Beem, J. S. Richardson, and D. C.Richardson. 1982. Determination and analysis of 2 A structure of copperzinc superoxide dismutase. J. Mol. Biol. 160:181-217.

van Aalten, D. M. F., A. Amadei, A. B. M. Linssen, V. G. H. Eijsink, G.Vriend, and H. J. C. Berendsen. 1995. The essential dynamics ofthermolysin: confirmation of the hinge-bending motion and comparisonof simulations in vacuum and water. Proteins Struct. Funct. Genet.22:45-54.

van Aalten, D. M. F., B. L. de Groot, J. B. C. Findlay, H. J. C. Berendsen,and A. Amadei. 1996. A comparison of techniques for calculatingprotein essential dynamics. J. Comput. Chem. 18:169-181.

van Gunsteren, W. F., and H. J. C. Berendsen. 1987. GROMOS: GroningenMolecular Simulation (GROMOS) Library Manual. Biomos, Groningen.

Viglino, P., A. Rigo, E. Argese, L. Calabrese, D. Cocco, and G. Rotilio.1981. '9F relaxation as a probe of the oxidation state of Cu, Zn super-oxide dismutase. Studies of the enzyme in steady-state turnover. Bio-chem. Biophys. Res. Commun. 100: 125-130.

Vriend, G. 1990. WHAT IF. A molecular modeling drug design program.J. Mol. Graph. 8:52-56.

Wong, C. F., C. Zheng, J. Shen, A. McCammon, and P. G. Wolynes. 1993.Cytochrome c: a molecular proving ground for computer simulations.J. Phys. Chem. 97:3100-3110.

Wong, Y., T. W. Clark, J. Shen, and J. A. McCammon. 1993. Moleculardynamics simulation of substrate-enzyme interactions in the active sitechannel of superoxide dismutase. Mol. Simul. 10:277-289.